2.5. Retrieval of land use/cover patterns and the temperature
Indices of NDVI, NDWI, NDBI and NDBaI are used in the
classification of images but it was difficult to determine the
appropriate threshold values for the indices to get ideal results
of classification. In order to solve this problem we analyzed the
relative spectral characteristics of different land use/cover types.
Here we take the Landsat 7 ETM+ 1G image on Nov. 1, 2000 as
an example for the spectral analyses. Through repeated
sampling of the representative pixels of each type, the values
in all seven bands were obtained and averaged (Fig. 3).
After the spectral characteristics of land use/cover types
were identified, Boolean operators were used on the spectral
bands to improve the results of indices. For example, an
operator (band4−band5) > 0 was combined with (band4−
band3) > 0 (or NDVI > 0) to extract vegetation from the Landsat imagery because only vegetation has the spectral feature of
band4−band3 > 0 and band4−band5 > 0. Similarly, water was
extracted by combining (band3−band4) > with (band4−
band5) > 0. We then calculated the intersection of (band3−
band4) > 0, (band4−band5) > 0, and NDWI >k(kis an estimated value to derive water areas) to retrieve all water bodies
in the study area. Then band ratio (band5/band2) was used to
dividethetotalwaterareaintoopenwatersurfaceandfish
pond (shallow water areas) by setting the appropriate thresh